ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper 분할된 휴대 수하물 데이터셋에 대한 다중 Pipeline과 결합한 다중 Student-Teacher 프레임워크의 효과에 관한 연구
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
정지욱, 송윤선, 이수열
Issue Date
2022-11
Citation
대한전자공학회 학술 대회 (추계) 2022, pp.2557-2561
Publisher
대한전자공학회
Language
Korean
Type
Conference Paper
Abstract
In this paper, we investigate the student-teacher framework involving multiple mean teachers coupled with the multiple pipelines on the partitioned SIXRay-10 dataset. Firstly, SIXRay-10 dataset is partitioned and the multiple pipelines are applied with ResNet-50 backbone and Faster R-CNN detector to get the enhanced detection accuracy of 80.6 mAP0.5. Secondly, for each pipeline, a mean teacher model is defined as an exponential moving average of the student model parameters and the pipelines are dynamically aligned according to its bounding box regression loss to get 82.1 mAP0.5.
KSP Keywords
Bounding Box, Coupled with, Detection accuracy, Exponential Moving Average(EMA), Faster r-cnn, Model parameter, Student model, Student-teacher, Teacher Model